TUMSAT-OACIS Repository - Tokyo University of Marine Science and Technology (東京海洋大学)
A Compositional Approach to Japanese Anaphora
journal or
publication title
東京商船大学研究報告. 人文科学
volume
50
page range
29-73
year
1999
URL
http://id.nii.ac.jp/1342/00000590/
A COMPOSITIONAL
APPROACH
TO JAPANESE
ANAPHORA
Masaaki Fuji
Tokyo University of Mercantile Marine
In this article I argue that Japanese complex anaphors such as kare-zisin or zibun-zisin are best analyzed as a DP, where the head zisin takes an NP complement such as kare or zibun. Syntactic evidence is presented to show that zisin is a D and kare or zibun is an NP. It is then shown that the DP analysis naturally explains various properties of the complex anaphors, including the fact that contrary to what has been observed in the literature, complex anaphors can take an antecedent outside their local domain.*
keywords: complex anaphors, binding theory, Japanese anaphora, DP 1. Introduction
Since Pica (1985, 1987), it has been claimed by many researchers that there are basically two types of reflexives in natural language, i.e., simplex and complex reflexives, and that a complex reflexive has an internal structure, just like an ordinary NP or DP. This basic idea, which I will call the compositional approach, has been developed by Reinhart and Reuland (1991, 1993), Katada (1991), among others.1 I believe their basic insight is correct, but their actual analyses still rely partially on residues of the traditional non-compositional approach. What I will try to do in this paper is bring their basic idea to the extreme. In other words, I would like to take the idea of compositionality at its face value, and pursue the theory in which the properties of the whole are completely derived from the properties of the parts. If this extreme position is successful, we need not stipulate any special principles exclusively for the whole, a welcome result.
To see how the previous analyses fail to execute compositionality in a full-fledged manner, let us take Katada's (1991) theory and examine how she treats Japanese reflexives such as kare-zisin 'him-self. Katada assumes that kare-zisin is a phrasal reflexive composed of kare plus zisin, as shown in (1), and notices that kare-zisin inherits an important property from one of its parts, kare, namely, the property that it cannot take a quantifier as an antecedent, as shown in (2):2
(Katada 1991: 294)
* This paper is a slightly revised version of the paper written in August 1994 at Rutgers University as the first generals paper for my Ph.D candidacy. I would like to thank Jane Grimshaw, Vivian Deprez, Ken Safir, and Maria Bittner for their invaluable comments, advice, and encouragement. All remaining errors are mine. 1 For further details of various sorts of compositional approach, see Battistella (1989), Browning (1992a, 1992b), Cole, Hermon, and Sag (1990), Safir (1993a, 1993b), Tang, C-C J. (1989), Tang, D-W (1989), and Yu (1992). 2 1 will discuss this peculiar property otkare in 4.3.1.3.
(30) M JEW
(2) a. *Darekai-ga [karerga katta to] omotta. someonerNom [hej-Nom won that] thought 'Someone, thought that he; won.' (Katada 1991: 305) b. *Dareka;-ga [kare-zisin]ro semeta.
someonej-Nom [he-self]rAcc blamed
'Someone blamed himself,.' (Katada 1991 : 300)
The above aspects of Katada's analysis is in accord with the spirit of compositionality. However, she treats kare-zisin as a single unit when computing its local domain, and seems to apply Condition A to the whole phrase. I have just used 'seems,' because she never explicitly mentions Conditions of the Binding Theory anywhere in her paper, but I suspect that she implicitly assumes at least something like Condition A in her framework. Otherwise, she could not explain why reflexives like zibun or kare-zisin must have an antecedent in its local domain. If my interpretation of her treatment of kare-zisin is correct, then we can conclude that even under her compositional approach, she sometimes has to treat a complex reflexive in a holistic manner, disregarding its internal structure. Once we admit that sometimes complex reflexives must be treated as a unit, we are forced to list them in the lexicon. This renders all commonalities between kare as a pronoun in an NP domain and kare in kare-zisin just a pure coincidence. Put differently, under this particular version of compositional approach, we must abandon one of its most important feature, i.e., the principle of compositionality, which states that the properties of the whole constituent are completely derived from the properties of its parts, and no specific principles need not be stipulated exclusively for the whole. On the other hand, under the theory which faithfully incorporates the principle of compositionality, we need not list kare-zisin in the lexicon, and all of its properties are derivable from its parts. This explains the existence of commonalities between kare in an NP domain and kare in kare-zisin, because both are just two occurrences of the same lexical item kare.
The organization of the paper is as follows. Section 2 establishes the internal structure of Japanese complex reflexives. Section 3 offers theories on anaphoric dependency which I will rely on in tin's paper. In Section 4, 1 will justify our version of compositional approach, arguing that what is called a 'complex reflexive' is just a syntactic phrase whose head is D°. Section 5 draws some conclusions.
2. Two Types of Anaphors 2.1. Categorial Status of Pronouns 2.1.1. XP-Pronouns and X°-Pronouns
Hestvik (1992) claims that natural language has two types of pronouns: XP-pronouns and X°-pronouns. XP-pronouns are the pronouns that are immediately and exhaustively dominated by a maximal projection, and X°-pronouns are the pronouns that project X-bar structure completely. The former is exemplified by English pronouns such as he, she, and it, and the latter, by Norwegian pronouns such as han 'he', and hun 'she'. The structures Hestvik (1992) proposes for both types of pronouns are the following:
(3) a. XP-pronouns (English) b. X°-pronouns (Norwegian)
NP NP
h
im N' XP
N (restrictive modifier) han
The above structures predict that X°-pronouns can take a restrictive modifier, but XP-pronouns cannot, because in the latter case, by definition, all and only material that NP's can dominate is a pronoun, and nothing else. This prediction is confirmed by the following observation by Hestvik (1992: 569):
(4)...[I]magine being in the situation of identifying a mugger in a police lineup. The offender is wearing a red hat. You say to the police officer next to you, without nodding or pointing, the Norwegian expression corresponding to "It's him with the redhat," or "He with the red hat is the guilty one," or "It was he who has a red hat that did it." This is ungrammatical in English, but perfect in Norwegian.
(Hestvik 1992: 569, Note 10) The most interesting consequence of this approach is that antisubject orientation exhibited by Norwegian pronouns and the lack thereof in English can be reduced to the X-bar theoretic differences in pronouns of each language, provided that pronouns, in addition to reflexives, can move at LF. Hestvik proposes the following requirements on movement of pronouns and reflexives:
(5) At LF,
a. X°-pronouns and X°-reflexives mustoccur in a functional head.
b. XP pronouns and XP-reflexives must occur in the Specifier of their governor.
(Hestvik 1992: 566)
The requirement in (5a) forces Norwegian pronouns to move from their D-structure position to the nearest head position of a functional category. Take, for example, sentences such as the one given in (6a), where hans cannot corefer to the subject:
(6) a. *Johnj liker [hanSi kone].
John; likes [his; wife]
b. Norwegian LF representation IP NP,
r
蝣
I
I
V P
蝣
hansi
V
D P
I
liker
N P
D '
N -
D
N P
t.
kon
e
(Hestvik 1992: 571)In (6b), hans, being an X°, moves to the nearest functional head, I, satisfying the requirement in (5a). Hestvik assumesthat pronouns must satisfy Condition B both at S-structure and LF. Hans in (6b) violates Condition B, because at LF, its governing category is the matrix IP, in which hans is bound by the subject NP.
Howeverelegant it may look at first, Hestvik's approach has both conceptual and empirical problems. First, there is one serious conceptual problem: the existence of XP-pronouns is against any versions of the X-bar
theory. This can be understood in the context of die development of the X-bar theory. What researchers have tried to eliminate is exactly the type of exception to the X-bar theory Hestvik introduces, namely excocentricity or headlessness. The categories S, S', and DetP were considered to be headless, which is against one of the defining properties of the X-bar theory, endocentricity or headedness. Consider the following structures:
( 32)
m sm
b.In (7a), DetP is headless. This is just a stipulation to capture the fact that determiners cannot be restrictively modified. Also in (7b), S' and S dominate no heads from which they are supposed to be projected. To eliminate this type of exception, syntactic categories like IP, CP, and DP have been introduced.3 Now, it is clear that Hestvik's introduction of XP-pronouns is conceptually undesirable, because it amounts to weakening the X-bar theory once again.
Second, Hestvik's analysis makes a wrong prediction in the case of Japanese pronouns. Japanese pronouns like kare 'he', and kanozyo 'she' can be used with restrictive modifiers. This is shown by the fact that the sentences in (8) can be uttered in the same situation as depicted in (4):
(8) a. [akaiboosi-no kare]-ga han'nin desu. [red hat-Gen /ze]-Nom criminal be 'He with a red hat is a criminal.'
b. sono hanzai-wa [akai boosi-o kabbuta kare]-ga. yarimasita. that crime-Top [red hat-Ace wear fce]-Nom committed "That crime, he who wears a red hat committed.'
According to Hestvik, this clearly indicates that Japanese pronouns are of X°-type. This predicts that Japanese pronouns must show anti-subject orientation, just like Norwegian pronouns. However, this prediction is not borne out, as shown in (9), where the pronoun kare and the subject John can be coreferential:
(9) Johnrwa kare^-no tuma-o aisite-iru. JohnrTop he^-Gen wife-Ace love-be 'Johni loves his^ wife.'
In sum, the conceptual and empirical problems just discussed make Hestvik's approach less attractive than it may look at first.4
3 See Chomsky (1986) for further discussion. There is also an alternative structure for NP, in which determiners do not project any bar levels:
(i) NP Dpt
boy
This analysis violates two principles of the X-bar theory. It is against a principle requiring a specifire to be XP. It is also against a principle requiring an X°-category to project to XP-level.
4 It is not clear to me why Norwegian pronouns show anti-subject orientation, while Japanese pronouns do not exhibit such orientation.
2.1.2. D-Pronouns and N-Pronouns
Since Postal (1966), it has been pointed out that there are similarities between pronouns and determiners. Abney (1987) claims that those similarities should be explained by assigning them the same functional category D.
Noguchi (1993) proposes that there are two types of pronouns in natural language: D-pronouns and N-pronouns. D-pronouns are the pronouns that head D-projections, while N-pronouns are the pronouns that liead N-projections. The former is exemplified by English pronouns such as he, she, and it, and the latter, by Japanese pronouns such as kare 'he', and kanojo 'she'. The structures that Noguchi (1993) assumes for these two types of pronouns are given in (10):
(10 ) a . D -p ro n o u n s (E n g lish ) b . N -p ron o un s (Jap a n e se )
D P N P
蝣
D l N l
D N P N
h e /sh e/it k are /k an o z y o
As one piece of evidence for the N status of Japanese pronouns and for the D status of English pronouns, Noguchi (1993: 9) directs our attention to the fact that English pronouns generally do not cooccur with a prenominal modifier, while Japanese counterparts can. This follows if we make a not unreasonable assumption that modifiers can only attach to N-projections, but not D-projections.
(ll) a. tiisanakare c. *smallhe small he
b. sinsetuna kanozyo d. *kindshe kind she
Noguchi's point can be strengthened by observing the following data: (12) a. kyonen-no kare c. *lastyear'she
last.year-Gen he 'what he was last year'
b. kinoo-no kanozyo d. *yesterday's she yesterday-Gen she
'what she was yesterday'
e. T a r o - w a [Tokyo-no kare] kara tegami-o morratta. Taro-Top [Tokyo-Gen he] fromletter-Ace received 'He received a letter from Tokyo's him.' (Lit.)
f. boku-ga moo-itido aitai no-wa [kyonen-no kare] da. I - N o m once.again want.to.meet Comp-Top [last.year-Gen he] be " W h oI want to meet once again is last year's him.' (Lit.)
The data in (12) indicate that Japanese pronouns can be preceded by possessives, but English pronouns cannot. This difference comes from the interaction between the N/D status of pronouns and the difference in where possessives are generated in each language. I assume with Abney (1987) that English possessives are generated in the Spec of DP, to which the genitive Case is assigned by AGR in D. This predicts that if the head of DP is occupied by some other element than AGR, possessives are never licensed. This assumption, together with
(34) ® JEW
Noguchi's claim that English pronouns are D's, explains why (12c) and (12d) are ungrammatical. This is because in (12c) and (12d), a pronoun, occupying the head of DP, excludes AGR, and without AGR, yesterday's is not licensed.
Japanese possessives, on the other hand, behave more like adjectives than like English possessives in that they can freely interchange with other adjectives. Compare (13) and (14):
(13) a. kyonen-no koofukuna kare last.year-Gen happy he b. koofukuna kyonen-no kare
happy last. year-Gen he (14) a. Tom'sbeautifulhouse
b. *beautiful Tom's house
I assume that interchangeability of the kind shown in (13) is a typical property of modifiers to N projections. As Fukui (1986) points out, prenominal adjectives in English basically exhibit this type of interchangeability. Though some semantic restriction on the ordering of adjectives make some orderings sound odd, scrambling among prenominal adjectives produces much better combinations than scrambling among all the prenominal elements including a determiner. Observe the contrast between (15) and (16):
(15) a. the tall, dark, handsome stranger b.??the tall, handsome, dark stranger c. ?the dark, tall, handsome stranger d.???the dark, handsome, tall stranger e.???the handsome, tall, dark stranger f.???the handsome, dark, tall stranger (16) a. *tall, dark, handsome, thestranger
b. *tall, the, dark, handsome stranger c. *dark, tall, the, handsome stranger d. *handsome, the, tall, dark stranger
The data just given follows form the descriptive generalization in (17), which I assume is applicable universally: (17) Syntactically, scrambling among prenominal modifiers to N projections is permitted, while prenominal
modifiers to N projections cannot move crossing elements generated within D projections.
Recall that in (13), a possessive and an adjective are allowed to be scrambled. This fact, together with the generalization (17), indicates that Japanese possessives are in fact modifiers to N projections, rather than specifiers ofDP. I want to claim here that if this is the case, then the data given in (12a) and (12b) constitute one piece of evidence for N status of Japanese pronouns. In (12a) and (12b), possessives are allowed to precede pronouns. We have already shown that Japanese possessives are modifiers to N projections. Therefore, it must be the case that Japanese pronouns are N's, rather than D's.
There is another piece of evidence given by Noguchi (1993) that shows that English pronouns are determiners and Japanese counterparts are nouns: the former can take an NP complement, while the latter cannot. Note that Japanese is strictly head-final, and hence we have to check combinations such as those given in (19), where nouns precede pronouns.
(18) a. usthreemen b. we Americans
c. you foolish soldiers (19) a. *nihonzin watasitati
Japanese we b. * keisatukan anatatati
policemen you
Noguchi (1993: note 8) also notices that "Japanese pronouns are inflected for number in the same way that commonnouns are," and suggests that "this parallelism can be taken as another indication that Japanese pronouns are N's."5 (20) a. otoko-ra b. kodomo-ra man-pl child-pl 'men' 'children ' (21) a. kare-ra b. kanozyo-ra he -pl she-pl
'they (male)' 'they (female)'
2.2. Zisin as a D-Anaphor and Zibun as an N-Anaphor
I propose thatjust like pronouns, there are two types of anaphors: D-anaphors and N-anaphors: zisin is a D-anaphor and zibun is an N-anaphor.
(22) Two Types of Anaphors in Japanese
Zisin heads a D-projection, and zibun heads an N-projection.
I assume zisin takes an NP complement, while zibun cannot. Thus the structures in which the two types of anaphors appear are such as those in (23):6
(23) D-anaphor N-anaphor DP NP S pec N1 N zisin zibun
The structures given in (23) correctly predict that zisin can take zibun, but not vice versa, because zibun itself is an NP and can be a complement to zisin. In the following, I will use 'selfN' and 'selfD' as a gross for zibun and zisin, respectively:
5 The affix -ra is different from English -s in that John-ra means a group of people which is characterized by John being a salient member of that group, rather than a group of people each member of which happens to be
named 'John.'
6 The fact that zisin takes an NP complement but zibun cannot may be due to one or both of the following two reasons: (1) zisin, but not zibun, assigns a special 0-role; (2) every functional head must take a specific complement, and thus zisin, but not zibun, must take a complement. In this paper, I want to claim that (2) is derived from (1).
( 36) JEW (24) a. [Dp[pzibunzisin]] s elfN selfD b. *zisin zibun selfD selfN
They also predict that zisin, but not zibun, can take various types of NP's. In particular, notice that the fact that zisin can take pronouns like kare and kanozyo corroborates Noguchi's (1993) claim that Japanese pronouns are NP's:
(25) a. [sono zyosei] zisin [that woman] selfD b. [tiisanakare] zisin
[short he] selfD c. [utukusii kanozyo] zisin
[beautiful she] selfD d. [orokana zibun] zisin
[foolish selfN] selfD
(26) a. *[sonozyosei] zibun [that woman] selfN b. *[tiisanakare] zibun
[short he] selfN c. *[utukusii kanozyo] zibun
[beautiful she] selfN d. *[orokana zibun] zibun
[foolish selfN selfN
There is another piece of evidence showing the D status of zisin, and the N status ofzibun. Zibun can be modified by an adjective, or a possessive, but zisin cannot:
(27) a. tiisanazibun short selfN
b. kinoo-no zibun yesterday-Gen selfN
'
what self was yesterday'
(28) a. *tiisana zisin short selfD
b. *kinoo-no zisin yesterday-Gen selfD
Further evidence for our analysis comes from number inflection. Zibun, but not zisin, is inflected in number in the same way as common nouns:7
7 In a later section, I will claim that zibun lacks 0-features as a lexical property. This claim seems to be inconsistent with the fact that the N-anaphor zibun can take a plural morpheme -ra, since this fact appears to indicate that zibun has at least the number feature. But I want to claim that zibun's ability to be affixed by -ra is independent of the lack or presence of ^-features in the lexicon. In the first place, zibun does have 0-features at LF, because as I will claim later, zibun, lacking (^-features as a lexical property, gets them from its antecedent at LF. Hence, in principle, -ra could be the realization of this newly acquired number feature. However, facts are more complicated than this. Observe the following data:
(i) a. Tarorwa zibunrga bakada-to omotteiru. TarorTop selfNi-Nom fool.be-that think
'TarOj thinks that self; is a fool.'
b. Tarorwa zibunj-ra-ga bakada-to omotteiru. TarorTop selfNi-ra-Nom fool.be-that think 'TarOj thinks that selfj.and others are fools.'
(ib) shows that the plural affix -ra in zibun-ra has nothing to do with the number feature passed to zibun from Taw, i.e., [-plural]. If the affix -ra had to be licensed by the feature [+plurral] on the stem zibun, then (ib) would not bejudged as a grammatical sentence. This is because the stem zibun in (ib) is marked as [-plural] rather than [+plural]..
(29) a. zibun-ra selfN-pl b. * zisin-ra
selfD-pl
Also, note in passing that not only pronouns but also proper names are N's in Japanese, for Japanese propernames pass all the criteria we have used to show the nominal status of a lexical item: (i) they can be a complement to zisin [(30a)], (ii) they can be preceded by prenominal adjectives and possessives [(30b,c)], and (iii) they are inflected in numberjust like common nouns [(30d)]:8
(30) a. Taro zisin Taro selfD b. utukusii Hanako beautiful Hanako c. kyonen-no Hanako last.year-Gen Hanako 'what Hanako was last year' d. Hanako-ra
Hanako-pl
3. Theories on Anaphoric Dependency 3.1. Spec-Head Agreement as Condition A
In this section I will depart radically from the previous literature and argue that the effects of Condition A on lexical anaphors are derivable from Spec-Head agreement.
3.1.1. The Feature System of Nominals
Before going into the detailed discussion of the reduction of Condition A to Spec-Head agreement, I will make several assumptions about the feature system of nominals.
First, I adopt Grimshaw's (1991) analysis of the syntactic categories, in which N and D have the same feature[+N].
Second, I adopt Hoji's (1990) feature analysis of [+ N] categories: (31) Features for [+ N] categories in Japanese
a. Anaphors: zibun ('selfN'), zisin ('selfD') [+a]
b. Pronominals: kare ('he'), kanozyo ('she'), sore ('it'),... [-a]
c. Epithets: yatu ('the guy'), aitu ('the guy'),... [-a]
d. Social Titles: sensei ('teacher'), daitoryo ('president'), [-a]
e. Names: Taro, Hanako, ringo ('apple'),... [-a]
Third, I assume that [-a] heads have ^-features, but [+a] heads do not.9 And I also assume as a part of the
definition of X-bar theory that features on the head percolate up to the maximal projection.
8 The NP Hanako-ra in (30d) does not mean the set of people whose names are all Hanako, but it stands for the set of people whose representative member is Hanako.
9 For die lack of ^-features, see Burzio (1991). I will follow Burzio (1991) and assume that anaphors lack
person, number, and gender features, but not Case feature. In the following, when I use ^-features for anaphors, I mean person, number, and gender features, excluding Case features.
(38)
m mm
3.1.2. The Theory of Spec-Head AgreementThe notions 'specifier' and 'head' in Spec-Head agreement are usually defined on the basis of the basic X-bar configuration given in (32a):
Thus, X° in (32a), which is defined as the head in the X-bar theory, also acts as the head in Spec-Head agreement. The crucial defining property of a head is that a head projects a single bar and a double bar levels. According to this definition, the adjoined element W° in (32b) is not the head of X' or XP, because it does not project any bar levels. A problem arises when we consider the following adjoined structures, where T and V are raised and adjoined to Agrs and Agr0, respectively:
In each case, the Case feature on the adjoined element mustbe in the Spec-Head relation to the specifier of AgrP. But the definition of a head given above disallows die adjoined element to be the head of AgrP, and hence the adjoined element cannot have the Spec-Head relation to the specifier of AgrP, i.e., DP.
Two possible solutions to this problem have already been suggested in Chomsky (1993). The first solution stipulates that the Case feature on T and V can percolate up to the upper segment of Agrs and Agr0, respectively. This stipulation makes it possible for the Case feature on the adjoined element to be checked off against the Case feature on DP through the conventional Spec-Head relation between DP and Agr. In other words, the checking relation between DP and T or V is established via the intermediate Agr.
The second solution is based on the set of newly defined domains. Among them, a checking domain is relevant to our problem. Consider the configuration (34). H is a zero-level category which raises and adjoins to X, forming the chain CH = (H,r). X is the head of X' and XP. The maximal projections ZP and YP are the specifier and the complement, respectively:
Chomsky assumes that only the chain CH = (H,t), rather than H itself, can have the domains, and both CH and the two-segment category X are defined to have the same checking domain, namely, ZP. If we assume that X is Agr0 and H is V, or that X is Agrs and H is T, then the (/"-features on ZP can be checked off against those on X by virtue of ZP being in the checking domain of X, and the Case features on ZP will also be checked off against those on CH by virtue of ZP being in the checking domain of CH.
I will adopt the second approach in this paper, not only because the domains introduced in the second approach are needed in several important cases, but also because only the second approach, but not the first, is in accord with our analysis of zibun binding.
With the above discussion in mind, let us move on to the theory of Spec-Head Agreement (SHA).10 I assume that SHA consists of two subprinciples: Feature Checking and Feature Passing. This means that if some projection is required to salisfy SHA, it can satisfy SHA by satisfying Feature Checking or Feature Passing. I give the definitions of these subprinciples in (35) and (36), where a set of features is represented as [F] and die lack thereof, as [ ]. Note that in both (35b) and (36b), [F] on X° percolates up to XP. XP is taken to be aandY°,B:
(35) a. FeatureChecking:
The features on an element (a) in the checking domain of B are, checked by the
features on B, and if they match, we say a satisfies Feature Checking.
(36) a. FeaturePassing:
The features on an element (a) in the checking domain offi pass onto./?, which lacks those features, and if this passing occurs, we say a satisfies Feature Passing.
Further, I assume the following licensing condition on [+N] maximal projections: (37) Feature Agreement Principle (FAP)
Every [+N] maximal projection must satisfy SHA.
This means that any NP or DP must satisfy either Feature Checking or Feature Passing.
Now, let me illustrate how die above mechanisms interact with each other to derive the effects of Condition A. Consider the following sentence with its S-structure and LF, where [</> F] represents a set of <j>-features, and [ ] stands for the lack of ^-features:
10 The idea that reflexives lack ^-features and their antecedent supplies them is first proposed by Burzio (1991), as far as I know. But the theory of Spec-Head Agreement to be proposed here is based on the idea suggested to me by Vivian Deprez (personal communication). For a similar approach to Spec-Head Agreement, see Deprez
(40)
m jew
(38) a. Taro-ga zibun-o kiratte-iru. Taro-Nom selfN-Acc hate-be b. S-structure AgrsP
Spec
Agrs ' Agrs [0F] AgroPS pec
Agr0 [0F] Agr0' NP VP V Taro NP V t«F] | | N' kiratteiru N zibun[ ] c. LF A grsP 蝣 I N P , A grs ' 蝣 I I T aro A grs A groP W F ] N , A grs N P , A gro' I [*F ] zibun N I A gro V P [ ] [* F ] N , N P , V t, N P , V t, k iratteiruAtLF, T a w m o v e sto the Spec ofAgrs andzibun moves to the Spec of Agr0 in order to satisfy SHA. At LF, T a w is in the checking domain of Agrs, which has ^-features. Hence, T a w satisfies Feature Checking. But zibun cannot satisfy SHA in the Spec of Agr0, because zibun has no ^-features and accordingly, the maximal projection, NPP has no ^-features to pass or to be checked. Hence, zibun has to move further to get ^-features. This time, zibun moves as an X° to adjoin Agrs. As I have assumed above, in this adjoined position, the chain CH = (zibun, t2) has the Spec of Agrs, NP3, in its checking domain, and thus it gets (^-features from NP3 by Feature Passing. I also assume that the ^-features assigned to the chain are shared by each member of the chain. This allows the tail of the chain, t2, to get the (/(-features. And the (/"-features on t2 percolate up to NP,, which is n o wable to take part in the Feature-Checking relation with Agr0.
If the story given above is correct, then we can dispense with Condition A, at least in the case of anaphor binding, and still account for the fact that zibun needs an antecedent. This is because zibun needs 0-features, and only way to get them is to move to Agrs, where zibun and its 'antecedent' in Spec of Agrs take part in the Spec-Head Agreement.
Further, w e need to explain why zibun has (i) the subject-orientation, and (ii) the capacity to take a long-distance antecedent. Here, I have to be satisfied with simply making the following assumptions:
(39) a. In Japanese, the Agrtowhich V is adjoined (i.e., Agro) cannottakepartin the Feature-Passing relation ,
b. In Japanese, Feature-Passing is optional.
(39a) prohibits zibun to adjoin Agr0, and effectively forces zibun to adjoin Agrs. This gives us the subject-orientation. (39b) makes movement of zibun non-local, because if the closest Agro to the base position of zibun does not participate in Feature-Passing, then zibun has to raise to the next higher Agro. Obviously, these conditions are just stipulations right now, but I hope that future study will show that they are derivable from more fundamental principles.
3.2. Condition B and Condition C
I will adopt Hoji's (1990) Conditions B and C of die binding theory to account for the properties of X-zisin. It is thus necessary to summarize his version of the binding theory before we go into the detailed analysis ofX-zisin.
First, I will repeat Hoji's feature system for each type of the [+N] expressions in Japanese: (40) Features for [+ N] categories in Japanese
a. Anaphors: zibun ('selfN'), zisin ('selfD') [+a]
b. Pronominals: kare ('he'), kanozyo ('she'), sore ('it'),... [-a]
c. Epithets: yatu ('the guy'), aitu ('the guy'),... [-a]
d. Social Titles: sensei ('teacher'), daitoryo ('president'),... [-a]
e. Names: Taro, Hanko, ringo ('apple'),... [-a]
Hoji's Conditions B and C are based on this feature analysis:
(41) Hoji's (1990) Conditions B andC
a. Condition B: A [-a] category mustbe free in its local domain.
b. Condition C: A [-a, -p] category mustbe free.
Hoji's (1990) definition of the local domain is the simplest one: the local domain for X is the minimal
NP or S that contains a subject and X. This definition suffices for his purposes, but not ours. Hence, I will
define the local domain by incorporating Chomsky's (1986) idea of the Complete Functional Complex (CFQ:
(42) a. A CFCofan argument ais adomain in which all thearguments ofthehead Bare realized, where B
is the head that assigns a 0-role to a.
(43) a. The localdomainfor ais theleastCFCofa.
b. agoverns Biff a m-commands B, and no maximal projection intervenes
between a and B.
c. YP
« i Y1
Letme illustrate how to define the local domain for or, using the tree in (43c). Suppose that Y is a one-place predicate which assigns an internal 0-role to a inside Y', and that a moves into Spec of YP for some reason.
(42) m JEBfl
Letme illustrate how to define the local domain for a, using the tree in (43c). Suppose that Y is a one-place predicate which assigns an internal 0-role to a inside Y\ and that a moves into Spec of YP for some reason.
This movement gives us a non-trivial chain whose head is a, and whose tail is tr Y corresponds to the head /? in the definition of the CFC, because Y assigns a 0-role to a (or more precisely, to the chain with a as the head and tt as the tail). Hence, the least CFC of a is YP. Consequently, the local domain for a is YP.
Hoji's Conditions B and C behave rather differently from the binding conditions we have been accustomed to. First, Condition B applies not merely to pronominals, but also to names (and other [-a] categories). Second, Condition C never applies to Japanese [-a] categories, let alone anaphors and pronominals. This modification of Conditions B and C is supported by the following observations made by Oshima (1979) and Hoji
(1990):
(44) Japanese names must be free in their local domain, but can be bound from outside of this domain. (44) can be exemplified in the sentences in (45):
(45) a. *[Johnrwa Johnro bengosita]. [JohnrTop JohivAcc defended] 'John; defended Johiij.'
b. Johnrwa [Johnrno hon]-o motte kita. JohnrTop [JohnrGen book]-Acc bring came 'John; brought John's book over.'
In (45a), the second occurrence of John is bound in its local domain, namely the whole sentence, and this sentence exhibits violation of Condition B. On the other hand, the second occurrence of John in (45b) is free in its local domain, namely the NP containing the second occurrence of John, but it is bound from outside of its local domain by die subject John, and so Condition B is satisfied in (45b).
3.3. Linking Theory and the Condition on Linking
Hoji (1990) proposes that in addition to the binding conditions, we have to assume the condition on linking (CL), which regulates linking relations between two NPs, and it refers not to features such as [+/- a], or [+/- p], but rather to referential hierarchy. This proposal is based on Lasnik's (1989) generalization, which I will call Condition D: u
(46) Lasnik's Generalization (Condition D)
A less referential expression may not bind a more referential one.
Hoji's CL is different from Condition D in that it constrains the possible linking, rather than the possible binding. This modification is motivated by the fact that Condition D effects can be suspended. Before
illustrating CL, I will first introduce Condition D and show how it works, for I believe it helps understand the intuitive content of the actual condition, i.e., CL.
In order to illustrate how Condition D works, we need to know the referential hierarchy, in which four types of NPs are ranked according to the referentiality:
(47) The Referential Hierarchy: A > B: A is more referential than B Name > Social Title > Epithet > Pronoun
Japanese has the following types of NPs, which participate in the referential hierarchy:
(48) Names: Taw, Hanako, gakusei ('student'), ringo ('apple'),...
Social titles: sensei ('teacher'), daitooryoo ('president'),... Epithets: yatu ('the guy'), aitu ('the guy'),...
Pronouns: kare ('he'), kanozyo ('she'), sore ('it'), pro,...
Now let me illustrate how Condition D works by considering the coreferential possibility between a pronoun and a name. Observe that in (49), only (d) violates Condition D, because in (49d), the less referential element kare binds the more referential element Taw, thereby violating Condition D:
(49) a. Tarorga [Tarorno haha]-o aisiteiru. TarorNom [TarorGen mother]-Acc love 'TarOj loves TarOi's mother.'
b. Tarorga [karerno haha]-o aisiteiru. TarorNom [hej-Gen mother]-Acc love 'TarOj loves his, mother.'
c. karerga [karerno haha]-o aisiteiru. herNom [herGen mother]-Acc love
'Hej loves hiSj mother.'
d. *karerga [Taro,-no halia]-o aisiteiru. herNom [TarOi-Gen mother]-Acc love
'Hej loves Tarofs motlier.'
Recall that as we assumed in 3.2., languages like Japanese do not have nominals with features [-a, -p], and hence the effect of Condition C never shows up in such a language. This explains the acceptability of (49a).
Instead of Condition D, Hoji (1990) introduces the rule of linking (RL) and the condition on linking (CL) to capture basically the same generalization as Condition D is designed to capture:
(50) The Rule of Linking
IfX and Y are coindexed and X is less referential than Y, X must be linked to Z (Z may be Y itself.) where:
(i) Z is more referential than or equally referential to Y, and (ii) Z is coindexed with X and Y.
(51) The Condition on Linking
If A c-commands B, A cannot be linked to B.
Let me illustrate how RL and CL work by using (49d). Let us take kare and Taw as X and Y, respectively. In (49d), kare and Taw are coindexed, and kare is less referential than Y, namely Taw. Hence, kare must be linked to Z. Let us assume that Z is Y in this case, which means that Z is also Taw. This equation is justified, because Taw, as Z, is equally referential to itself, and Taro is coindexed with itself. Therefore, RL requires that kare must be linked to Taw in (49d). But this linking is prohibited by CL, because kare c-commands Taro. Therefore, RL and CL together predict (49d) is out.
This linking approach makes a different prediction from Condition D. It predicts that the so called Condition D effect is suspended in the following situation:
(44) M JEW
(52) The Suspension of the Condition D Effect
... Yt... Z(Name) > Y(Pronoun) > X(Social tiUe)
In (52), X and Y are coindexed and X is less referential than Y. Hence, X must be linked to Z or Y. In this case, X cannot be linked to Y, because this linking is banned by CL. Hence, X must be linked to Z. Therefore, the linking approach predicts that this configuration is ruled in. On the contrary, Condition D predicts that (52) is ungrammatical, because in (52) the less referential element X binds the more referential element Y.
Relevant data favor the linking approach, as can be seen from the following sentences: (53) a. [Matumoto sensei]rwa [kanozyorga [[Taro-ga senseij-no ie-made todoketa]
[Matumoto teacher]rTop [sherNom [[Taro-Nom teacherrGen house-to delivered] repooto]-o nakusitesimatta] to omotteita.
report]-Acc lost] that thought
'Ms. MatumotO; thought that shq lost the report that Taro had delivered to the teacher's house.'
cf. b. *ZirOj-wa [kanozyorga [[Taro-ga sense^-no ie-made todoketa] ZirOj-Top [sherNom [[Taro-Nom teacherj-Gen house-to delivered] repooto]-o nakusitesimatta] to omotteita.
report]-Ace lost] that thought
'ZirOj thought that she, lost the report that Taro had delivered to the teacher/s house.'
In (53a), we may take Matsumoto sensei as Z, kanozyo as X, and sensei as Y. Then, the linking approach predicts that (53a) is grammatical, because X has a more referential antecedent above it, namely Matsumoto sensei. Further, the linking approach correctly predicts that (53b) is out, because in this case, the subject NP Ziro cannot be taken as Z, because it is not coindexed with kanozyo. The binding approach makes a wrong prediction in the first case, since a less referential element {kanozyo) binds a more referential element (sensei) in (53a), thereby inducing Condition D violation.
It is now obvious that the relevant facts are more properly handled by the linking theory than Condition D. Therefore, I will henceforth use the linking approach in the following discussion.
4. A Compositional Analysis of X-Zisin 4.1. X-Zisin as the Local Domain for
X-The purpose of this section is to show that given the definition of the local domain introduced in 3.2., X-zisin itself becomes the local domain for X. I will repeat the definitions of CFC, local domain and government
in(54):
(54) a. A CFCofan argument ais adomain in which all the arguments oftheheadBarerealized, where B is the head that assigns a 0-role to a.
b. Thelocaldomainfor ais theleastCFCofa.
c. agoverns Biff a m-commandsB, and no maximal projection intervenes between a and B.
Let me illustrate how (54) works by using kare-zisin as an example. As we will see in 4.2., kare in kare-zisin is first generated as a complement to zisin, and then raises to the Spec of DP, as can be seen in (55):
(5 5 ) a . S -s tru c tu re b . L F D P D P D l N P D ' N P D N l N P D N ' z is in N t. z is m N k a r e , k a r e
I assume that Binding Theory applies at LF. So, we have to determine the local domain for kare in the representation given in (55b). The local domain for kare is the least CFC of kare (see (54b)). Now, a question arises as to whether or not kare has a CFC in the first place. I believe the answer is positive. Abney (1987) argues that every functional head assigns a functional role (F-role), a kind of #-role, to its complement. Suppose Abney is correct. Then the functional head zisin also assigns a F-role to its complement kare. A CFC ofkare is a domain in which all the arguments of zisin are realized (see (54a)). In (55b) the DP is qualified as the domain in question. This is because this DP dominates both kare, the only argument of zisin, and its head zisin. And obviously this DP is also the smallest CFC of kare. Hence, the local domain for kare in (55b) must be the DP.
4.2. Zisin and Spec-Head Agreement
Like the N-anaphor zibun, die D-anaphor zisin, being a [+a] head, lacks ^-features. Hence, it has to get <p-features, which is needed for the maximal projection DP to satisfy FAP. In order to get ^-features, zisin has to be in the Spec-Head relation with the specifier which has (^-features to pass. I will claim that the complement NP to zisin raises to the Spec of DP, and passes 0-features to zisin. This movement is in accord with 'Greed,1 because if the complement NP does not raise, it violates FAP. Consider the following structures for Taro-zisin ('Taro-selfD') and kare-zisin ('he- selfD'):
(56) a. S-structure DP b . L F D P N P D ' N I N P D N t z is in [* F ] T a r o /k a re [ ｫ F ] Taro/kare [0F]
In (56a), Taro/kare cannot pass its ^-features to D, and so it violates FAP in situ. Hence, it raises to Spec of DP at LF, as shown in (56b). From this position, it passes its ^-features to zisin, satisfying SHA and FAP. Nowzisin has ^-features, and they percolate up to DP, which also needs ^-features to satisfy FAP. Note that in
(46) m EBfl
our system, we need not stipulate zisin has to get (^-features. This is derived from FAP. To see this last point, let us consider the following sentence and its LF representation:
(57) a. Ziro-ga karerzisin-o aisite-iru.
ZirOi-Nom hei-selfD-Acc love-be
'ZirOj loves himself,.' AgrjP N P , A g r s ' Z ir o A g rs A g ro P [ O F ] 1 <P F ] D P , A g ro N R D ' A g ro V P [ O F ] N ' N P 3 D V N z isin t , V [ <P F ] k a r e a is ite -ir u [ O F ]
Suppose (^-features do not pass from NP3 to D. Then, DP2 would end up having no ^-features, which is a violation of FAP, because Agr0 cannot check ^-features on DP2.
The theory presented so far predicts that DP's such as Taro-zisin or kare-zisin can be used without any 'antecedent,1 because zisin gets ^-features DP-internally, and hence zisin need not raise to Agrs to enter into the Spec-Head Agreement relation with NP,, a potential antecedent in the binding theoretic terms. This prediction is indeed borne out:12
(58) a. [Elisabeth zvoo.oo zisin]-ga kuruma-o unten-sita. [Elisabeth queen selfD]-Nom car-Ace drive-did 'Queen Elisabeth herself drove the car.'
b. [kare zisin]-ga boku-ni aini-kita. [he selfD]-Nom I-to meet-came 'He himself came to see me.'
4.3. X-Zisin and the Theories on Anaphoric Dependency
In this section I will pursue the consequences of the compositional approach by investigating how X in X-zisin behaves with respect to the theories on anaphora introduced in Section 3.
12 As the translation of the sentences in (58) indicate, this use of X-zisin induces an emphatic reading. I will adopt die informal definition of an emphatic reading proposed by Aikawa (1994).
(i) Informally, an emphatic reading of a referent X is such that the referent X is put forward or intensified, while other alternatives under consideration are being excluded for the agent of a proposition in question.
(Aikawa 1994: 28)
For instance, in (58a), the referent of Elisabeth zyoo-oo zisin 'Queen Elisabeth herself' is put forward or intensified, while other alternative individuals are being excluded for the agent of the act of having driven the car. See 4.3.1.2., where I will discuss more of the emphatic reading, and show that the holistic approach cannot handle the emphatic reading properly.
4.3.1. Kare-Zisin
4.3.1.1. Kare-Zisin and Condition B
As pointed out in 4.1., the local domain for tore- in kare-zisin is the DP kare-zisin itself. Given this, it is predicted that tore- in kare-zisin behaves exactly like kare- in kare-no hahaoya 'his mother' with respect to Condition B. In this section I will show that this is indeed the case.
First, as we have already shown in 4.2., kare- can be used without any antecedent in its local domain:
(59) a. [DPkare-zisin]-ga boku-ni aini-kita. [DPhe-selfD]-Nom I-Dat meet-came 'He himself came to see me.1
b. [NP kare-no hahaoya]-ga boku-ni aini-kita. [NP he-Gen mother]-Nom I-Dat meet-came 'His mother came to see me.'
Second, kare- has no subject-orientation, because kare- is allowed to coindex with any NP outside of its local domain:
(60) a. Tarorga Jirorni [kare^-zisinl-nituite hanasita. TarorNom JirOj-Dat [he^-seiy-about talked 'TarOi talked to JirOj about himselfj/j.'
b. Tarorga JirOj-ni [kare^-no hahaoya]-nituite hanasita. TarorNom JirOj-Dat [he^-Gen mother]-about talked TarOj talked to JirOj about his^ mother.'
Third, kare- can have an antecedent outside the smallest AGRsP containing kare-. This may be controversial, because it has been assumed that kare-zisin as a whole is an anaphor, and obeys the Specified Subject Condition, but not the Nominative Island Condition. This means that kare- cannot have an antecedent outside of the smallest AGRsP containing kare-, unless tore- itself is a subject. Katada (1991), for instance, cites the following sentences to make this point:
(61) a. Johnrga [Billj-ga Mike^m kare-zisin?.i/j/k-no koto-o hanasita to] itta. John,-Nom [Billj-Nom Mike^-Dat he-self?.i/J/lt-Gen matter-Ace told that] said 'Johni said that Billj told Mikek about he-self,.i/j/k.'
b. Johiij-ga Billj-ni [kare-zisinj/j-ga katta to] itta. Johiij-Nom Billj-Dat [he-selfyj-Nom won that] said
'Johni told Billj that he-self^ won.1 (Katada 1991: 289)
According to Katada (1991), kare-zisin in (61a) cannot take John as its antecedent, because kare-zisin is inside the embedded sentence with a Specified Subject, Bill. On the other hand, in (61b), kare-zisin is a subject of the embedded sentence, so that it can escape die effect of the Specified Subject Condition, and hence take any NP in
the matrix as a potential antecedent. Contrary to Katada (1991) and most of the researchers, I would like to claim that we should not generally rule out sentences like (61a) with kare- coindexed with the matrix subject, because we can easily construct a sentence in which kare- occupies an embedded non-subject position, and still may take the matrix subject as its antecedent:
(48) B JEW
(62) a. Johnrwa [Maryrga Susank-ni karei/.j/.k-zisin-no himitu-o barasita] JohnrTop [Maryj-Nom Susan-Dat he^^-selfo-Gen secret-Ace let.out] koto-o okotteita.
that-Ace was.angry
'Johiii was angry that Mar^ let out hisi/.j/.k own secret to Susank.' b. Tarorwa [minnaj-ga karejy.j-zisin-o kiratteiru] to omotteita.
TarorTop [everyonej-Nom heiy.j-selfD-Acc hate that thought 'TarOj thought tliat everyone; hated himself^.'
c. titipwa [musumej-ga karei/.j-zisin-o kiratteiru] koto-o fatlier-Top [daughterj-Nom hei/,j-selfD-Acc hate] that-Acc hazimete sitta.
for.the.first.time learned
'The father; learned for the first time that his daughter hates himselfi/.j.' d. Tarorwa [JirOj-ga karei/.j-zisin-o homete-kureta] to itteita.
TarOj-Top [JirOj-Nom he^-selfD-Acc praise-gave] that said 'TarOi said that JirOj praised himselfi/.j.'
e. Tarorwa [JirOj-ga karej/.j-zisin-ni tyokusetu situmon-sini-kita] koto-o TarorTop [JirOj-Nom he^-selfD-Dat directly question-ask-came] that-Acc yorokonda.
was.glad
'TarOj was glad that JirOj himself came to ask questions to himselfj/.j.'
The most crucial property shared by all the sentences in (62) is that inside the embedded clause, there is no possible antecedent for hare-. In (62a) there are two NPs that are structurally allowed to act as an antecedent for kare-, but neither of them can be the actual antecedent, because they do not agree in gender with kare-. In (62b) since kare- cannot take a quantified NP as its antecedent in general, minna 'everyone' cannot be the actual antecedent. The embedded subject musume 'daughter' in (62c) does not serve as an antecedent to kare-, because of the difference in gender. Also in (62d) and (62e) the referent of kare- may not be die embedded subject Jiro. I would like to claim, following Kuno and Kaburaki (henceforth, K&K) (1977), that this is caused by the interaction between -zisin and helping verbs like -kureru 'give' and -kuru 'come'. Before dealing with this particular case, it is necessary to introduce K&K's (1977) proposal on empathy and its interaction with syntax.
K&K (1977) defines the notion 'empathy' as shown in (63a): (63) a. Empathy
Empathy is the speaker's identification, with varying degrees (ranging from degree 0 to 1), with a person who participates in the event that he describes in a sentence.
b. John hithis wife, [his wife=Mary]
c. Mary's husband hit her. [Mary's husband= John]
For instance, in (63b), the speaker identifies himself with John rather than with Mary. On the odicr hand, in (63c), the speaker identifies himself with Mary rather than with John. In the following, I use '>' to indicate this kind of the empathy relationship. Namely, 'A > B' stands for 'the speaker identifies himself with A rather than with B'. So, the empathy relations in (63b) and (63c) can be expressed with this notation; (64a) and (64b) correspond to (63b) and (63c), respectively:
(64) a. John>Mary b. Mary>John
Now, let us turn to Japanese giving verbs. Japanese uses two different verbs to express the meaning of 'give': yaru and kureru. K&K (1977) claim that yaru is subject-centered, and kureru, non-subject-centered. In other words, yaru, as a subject-centered verb, is used "when the action is looked at from the point of view of the referent of the subject, and not from the point of view of the referent of the dative object [i.e. the non-subject: MF]" (K&K 1977: 630). On the other hand, kureru, as a non-subject centered verb, is used "when the action is looked at from the point of view of the referent of the dative object [i.e. the non-subject: MF], and not from the point of view of the referent of the subject (K&K 1977: 630)." To show the effect of the subject-, and non-subject-, centrality, they use the following sentences:
(65) a. Boku-wa Taro-ni okane-o yatta. (Subject-Centered) I-Top Taro-Dat money-Ace gave
'I gave money to Taro.'
b. * Taro-wa boku-ni okane-o yatta. (Subject-Centered) Taro-Top I-Dat money-Ace gave
'Taro gave me money.'
(66) a. *Boku-waTaro-ni okane-o kureta. (Non-subject-Centered) I-Top Taro-Datmoney-Ace gave
'I gave money to Taro.'
b. Taro-wa boku-ni okane-o kureta. (Non-subject-Centered) Taro-Top I-Dat money-Ace gave
'Taro gave me money.1 (K&K 1977: 631)
The empathy relationship in each of the sentences given above can be expressed as follows: (67) (65a): Speaker > Taro
* (65b): Taro > Speaker * (66a): Taro > Speaker (66b): Speaker > Taro
To account for the unacceptability of (65b) and (66a), K&K propose the following constraint: (68) Speech-Act Empathy Hierarchy (SAEH)
It is not possible for die speaker to empathize more with someone else than with himself.
In (65b) and (66a), as we have already seen in (67), the speaker empathizes more with Taro than with himself, thereby violating SAEH.
K&K further try to account for the unacceptability of (69b):
(69) a. Tarorwa [Hanako-ga zibunrni kasite-kureta] okane-o tukatte-simatta. TarorTop [Hanako-Nom selfrDat lending-gave] money-Ace spending-ended.up 'TarOj has spent all the money that Hanako had lent to hinv'
b. *Tarorwa [Hanako-ga zibuiij-ni kasite-yatta] okane-o tukatte-simatta. TarOj-Top [Hanako-Nom selfrDat lending-gave] money-Ace spending-ended.up 'TarOi has spent all the money that Hanako had lent to him;.'
(50) * JEW
They first propose the following constraint:
(70) The Ban on Conflicting Empathy Foci (BCEF)
A single sentence cannot contain logical conflicts in empathy relationships.
(K&K 1977: 632) They also make the following assumption on the empathy relationship of zibun:
(71) The Empathy Relationship of 'Zibun'
Japanese reflexive pronoun zibun 'self, as it is used in a subordinate clause of the type illustrated in [(69)], requires that the speaker empathize with its referent rather than with other persons that show up in the same clause.
Given (71) and the empathy relationships of the giving verbs, the empathy relationships that hold in the embedded clauses in (69) are as shown in (72):
(72) (69a): zibun: Taro Hanako
kureta: Taro > Hanako (Non-subject-Centered) * (69b): zibun: Taro > Hanako
yatta: Hanako > Taro (Subject-Centered)
(69b) is unacceptable, because the empathy relationship of zibun and that of yatta conflict with each other, violating BCEF.
Now let us consider the sentences we introduced in (62d) and (62e), repeated here as (73a) and (73b), respectively :
(73) a. Tarorwa [JirOj-ga karej/.j-zisin-o homete-kureta] to itteita. TarOj-Top [JirOj-Nom hei,.j-selfD-Acc praise-gave] that said 'Taro; said that JirOj praised himself,,.;.1
b. TarOj-wa [JirOj-ga karej/.j-zisin-ni tyokusetu situmon-sini-kita] koto-o TarorTop [JirOj-Nom he^-selfo-Dat directly question-ask-came] that-Ace yorokonda.
was.glad
'TarOj was glad that JirOj himself came to ask questions to himself^.' First, I would like to generalize (71) to (74):
(74) The Empathy Relationship of Japanese Reflexives
Japanese reflexives zibun 'selfN' and zisin 'selfD' , as they are used in a subordinate clause, require that the speaker empathize with their
referent rather than with other persons that show up in the same clause.
Second, I assume that kuru 'come' is a non-subject centered verb, just like kureru 'give'. This means that the speaker using this verb empathizes more with its non-subject than with its subject.
(75) (73a):
i) zisin=Taro: zisin: Taro>Jiro kureta: Taro > Jiro *ii) zisin=Jiro: zisin: Jiro>Taro
kureta: Taro > Jiro (73b):
i) zisin= Taro: zisin: Taro > Jiro kita: Taro > Jiro *ii) zisin=Jiro: zisin: Jiro>Taro
kita: Taro > Jiro
As is clear from (75), BCEF prohibits -zisin to have the embedded subject Jiro as its antecedent.
To recapitulate so far, we have seen that kare-zisin can have non-local antecedent, only when the local subject is unavailable as an antecedent for kare-. It has also been observed that this unavailability of the local subject stems from the three reasons; (i) mismatch of (^-features between kare- and its antecedent, (ii) kare-'s inability to take a quantified antecedent, and (iii) the pragmatic constraint on the conflicting empathy relationships. 13
4.3.1.2. Further Evidence for the Compositional Approach
I have not mentioned in the preceding section that the sentences in (62), where kare-zisin has a non-local antecedent, actually induce emphatic readings just like the sentences in (58). In this section, we will see that this fact leads us to an argument against the holistic approach and in favor of our compositional approach, with several auxiliary assumptions.
In order to achieve this goal, we first need to clarify the relationship between the syntactic and pragmatic components. It is plausible to assume that both of the components are autonomous, and that the LF representations are input to the pragmatic component. It is also plausible to assume that the pragmatic component plays a role of filtering out pragmatically undesirable representations, without ever saving the output from the syntactic component that has already been marked as ungrammatical.
Within the holistic approach, we can offer two different analyses of the fact that kare-zisin is interpreted either emphatically or non-emphatically. The first analysis assumes that Japanese lexicon contains just one kare-zisin, whose distribution is constrained by Condition A. Or alternatively, we may assume that in Japanese lexicon, there are two subtypes of kare-zisin: the emphatic kare-zisin and the non-emphatic kare-zisin. The first analysis, call it the Holistic Analysis with 1 Kare-Zisin (HA-1, for short), is untenable. To see this, let us consider the following:
(76) a. Johnrga [Billj-ga kare-zisin^-o aisiteiru]to itaa. JohnrNom [BillrNom he-selfD?.i/rAcc love] that said 'John; said that Billj loves himself,.^.'
13 It is possible for kare-zisin to have a non-local antecedent even if there is an intervening antecedent. This happens only when the intervening antecedent is also non-local (The local domain for kare-zisin is indicated by square brackets.) :
(i) Tanaka-moto-syushoo-wa sudeni taihosareteiru zibun-no hisyo-ga kensatukan-ni The.former.prime.minister.Tanako-Top already arrested selfN-Gen secretary-Nom prosecutor-Dat [kare-zisin-ga hanzai-ni kakawatteita] koto-o morasu-nodewa-naika-to simpaisiteiru.
[he-selfD-Nom crime-Dat involved] that-Acc let.out-be-may-that is.worried
The former prime minister Tanaka, is worried that his secretary^, who has already been arrested, may let out to the prosecutor the fact that heyi himself was involved in the crime.'
(52) m lEBjJ
b. Johnrwa [Billj-ga kare-zisini/.j-o homete-kureta] koto-o JohnrTop [Billj-Nom he-selfDi/.j-Acc praise-gave] that-Acc yorokonda.
was.glad
'John; was glad that Billj had praised himselfi/.j.'
Under HA-1, kare-zisin is constrained by Condition A in the syntactic component. Hence, kare-zisin in both of the above sentences must be bound by Bill, butnot by John, at the level ofLF. Put differently, in (76a), kare-zisin takes Bill, but not John, as its antecedent at LF. And in (76b), contrary to the fact, kare-zisin takes Bill, but not John, as its antecedent at LF. The schematic LF representation for (76b) is given in the following:
(77) Johnrwa [Billrga [kare-zisinVj-o... kureta]....
To obtain the desired result, we have to posit the following processes in the pragmatic component: (i) Bill aixJ kare-zisin must be made non-coreferential. (ii) John and kare-zisin must be made preferential. The first process is not problematic, because pragmatic processes are supposed to function as filters. But the second process is problematic, because its effect is to make acceptable the representation that has already been made ungrammatical
in the syntactic component. This is impossible in the pragmatic component. Therefore, HA-1 cannot explain (76b), and is thus untenable.
Now, let us move on to the second analysis, call it the Holistic Analysis with 2 Kare-Zisin's (HA-2, for short). Under this approach, each sentence in (76) has two different derivations, i.e., one with the non-emphatic kare-zisin, and the other with the emphatic kare-zisin (I will use KARE-ZISIN for this usage.). The two different LF representations for each of the sentences are given below:
(78) LF Representations for (76a)
a. LF Representation with Non-Emphatic Kare-Zisin Johiyga [Billj-ga [kare-zisinj^-o...]....
b. LF Representation with Emphatic Kare-Zisin
Johnrwa [Billj-ga [KARE-ZISINJ^-o...]....
(79) LF Representations for (76b)
a. LF Representation with Non-Emphatic Kare-Zisin
Johnrwa [Billj-ga [kare-zisin].i/ro... kureta]....
b. LF Representation with Emphatic Kare-Zisin
Johnrwa [Billj-ga [KARE-ZISIN]i/ro... kureta]....
In (78a) and (79a) kare-zisin is non-emphatic, and obeys Condition A. On the other hand, I assume that KARE-ZISINin (78b) and (79b), which is emphatic, does not obey Condition A. Instead, it obeys both Condition B and an optional condition which stipulates that KARE-ZISIN must be bound from outside its local domain.14
Let us now consider what happens in the pragmatic component. The following two pragmatic conditions seem to be relevant here:
14 The second condition needs to be optional, since KARE-ZISIN can be used without an antecedent: (i) KARE-ZISIN-ga kuruma-o untensita.
he-self-Nom car-Ace drove 'He himself drove the car.'
(80) a. The Banon Conflicting Empathy Foci (BCEF)
A single sentence cannot contain logical conflicts in empathy relationships.
(Kuno and Kaburaki 1977: 632) b. Emphatic Reflexive Condition (ERQ
An emphatic reading of a referent of X is licensed only when its non-emphatic reading is unavailable.
The representation in (78a) is allowed in the pragmatic component. But the representation in (78b) is ruled out by ERC, because the representation in (78a) licenses a non-emphatic reading of a referent of KARE-ZISIN, which makes an emphatic reading unavailable. On the other hand, the representation in (79a) is ruled out by BCEF, because in (79a) kare-zisin empathizes more with Bill, but kureta empathizes more with John. Hence, a non-emphatic reading of a referent of kare-zisin is unavailable. The representation of (79b) satisfies both ERC and BCEF, because a non-emphatic reading of a referent of kare-zisin is unavailable, and because both KARE-ZISIN and kureta empathize more with John. Therefore, among the representations in (78) and (79), only (78a) and (79b) are well-formed in the pragmatic component. This is consistent with the fact.
It is, then, tempting to conclude that HA-2 is correct, and thus the holistic approach is on the right track. However, there is a severe conceptual problem with HA-2. To see this, observe the following:
(81) a. (N.Y.-no) Johnrwa [(L.A.-no) Johnrga [John-zisin]?«i/j-o hihansiteita] (N.Y.-Gen) JohnrTop [(L.A.-Gen) JohnrNom [John-selfD],.j/j-Acc criticized] to itta.
that said
'John, in N.Y. said that John, in L.A. had criticized [John-selfc],.^.'
b. (N.Y.-no) Johnrwa [(L.A.-no) Johnrga [John-zisin]i,.j-o homete-kureta] (N.Y.-Gen) JohnrTop [(L.A.-Gen) Johnj-Nom [John-selfD]i/,j-Acc praise-gave] koto-o yorokonda.
that-Ace was.glad
'John; in N.Y. was glad that Johnj in L.A. had praised [John-selfD]i/.j.1
(82) a. Tanaka-senseij-wa [Suzuki-senseij-ga [sensei-zisin],,^-© hihansiteita]to itta. Tanaka-teacherrTop [Suzuki- teacherj-Nom [teacher-selfD]?.i/j-Acc criticized] that said 'Mr. Tanakaj said that Mr. Suzukij had criticized [teacher-selfo],^.'
b. Tanaka-senseij-wa [Suzuki-senseij-ga [sensei-zisin]i/.j-o homete-kureta] Tanaka-teacheri-Top [Suzuki-teacherj-Nom [teacher-selfD]i/.j-Acc praise-gave] koto-o yorokonda.
that-Acc was.glad
'Mr. Tanakaj was glad that Mr. Suzukij had praised [teacher-selfD]i/.j.1
Note that in (81) and (82) we obtain the same anaphoric pattern as we observed in (76). So under HA-2, we have to posit that the Japanese lexicon includes emphatic and non-emphatic subtypes of both John-zisin and sensei-zisin. If the list is limited to kare-zisin, John-zisin, and sensei-zisin, the redundancy may not be so problematic. But in fact, this list must include a huge number of lexical items whose form is X-zisin. For instance, we can replace John in John-zisin with any human names, and the resulting form can be substituted for John-zisin in (81). The redundancy is overwhelming, and therefore we can conclude that even HA-2 is untenable.
It is easy to see that our compositional approach is free from this problem. Under this approach, all we need to posit is that there are emphatic and non-emphatic zisin in the lexicon, and it is not necessary to specify that kare-zisin, John-zisin, and so on have two different subtypes. We call this approach the Compositional Approach with 2 zisin's (CA-2, for short). Obviously, CA-2 radically reduces the number of lexical items we
(54) m jo.*!
have to store in the lexicon. We can thus conclude that CA-2 is superior to HA-2, and therefore the compositional approach must be chosen over the holistic approach.
4.3.1.3. Kare-Zisin and Quantified NP Antecedents
Our compositional approach predicts that kare- in kare-zisin displays exactly the same properties as kare within an NP domain. We have already seen in the preceding section that this is indeed the case with respect to Condition B. Now let us move on to another property which is shared by kare- in kare-zisin and kare within an NP domain.
It has been noted that the Japanese pronoun kare, unlike the English pronoun he, cannot be interpreted as a variable bound by a quantified antecedent. This can be seen in (83):
(83) a. *daremOj-ga [karerga tensaida] to omotteiru. everyonerNom [hej-Nom genius be] that think be 'Everyonet thinks that hej is a genius.'
b. *darej-ga [karerga katta] to omotteiruno. whorNom [herNom won] thatthink be Q 'WhOj thinks that he; won?1
However, Hoji(1991) points out that the more referential a quantified antecedent is, the more acceptable the bound/coreferential interpretation of kare becomes. Compare the sentences in (84):
(84) a. *darerga [Mary-ga karero buttato] ittano. whOj-Nom [Mary-Norn herAcc hit that] said Q 'WliOj said that Mary had hit him,?'
b. ??[dono hito]j-ga [Mary-ga karero buttato] ittano. [which person]rNom [Mary-Nom herAcc hit that] said Q 'Which person said that Mary had hit him;?'
c. ?[dono sakka]rga [Mary-ga karero buttato] itta no. [which writers-Norn [Mary-Nom herAcc hit that] said Q 'Which writer; said that Mary had hit him;?'
d. [dono nooberusyoo-zyusyoo-sakka]rga [Mary-ga karero butta to] itta no. [which Nobel Prize-winning-autlior]rNom [Mary-Nom hq-Acc hit that] said Q 'Which Nobel Prize-winning author said that Mary had hit himj?'
Interestingly, if we replace kare with kare-zisin in (83) and (84), exactly the same pattern is obtained. This is a strong confirmation of our compositional approach. This is because if we do not equate kare- in kare-zisin with the pronoun kare within an NP, assuming kare-zisin is an anaphor and kare a pronoun, then there is no obvious reason why they behave exactly the same when they are forced to be construed with a quantified NP antecedent with a varying degree of referentiality. Put differently, the fact that the subtlety of the judgments reported in (84) is reproduced with kare-zisin can only be explained by the compositional approach. Note, in passing, that the use of the term 'explain' here is appropriate. What we are trying to 'explain' is not the fact that kare tends to resist being construed with a quantified NP antecedent, but the fact that kare- in kare-zisin and kare within an NP behave in exactly the same way. The latter fact indeed follows from, and hence is explained by, our compositional approach.
4.3.2. pro-Zisin
4.3.2.1. Stylistic Conditions on the Usage of pro-Zisin
At first sight, zisin seems to be a bound morpheme, because as we have already seen, zisin is typically
used with a complement NP [(85)], but does not seem to be used as an independent word, without an overt
complement [(86)]:
(85) kare-zisin, kanozyo-zisin, sore-zisin, zibun-zisin, Mary-zisin, etc.
he-selfD she-selfD it-selfD selfN- selfD Mary-selfD
(86) a. Tarorga ??zisinj/zibunro nagutta.
TarOi-Nom self^/self^-Acc hit
'TarOi hit himself;.'
b. Tarorga ??zisini/zibunro kenasita. TarorNom self^/self^-Acc despised 'TarOj despised himself;.'
c. TarOj-ga ??zisini/zibunrno kokyoo-ni kaetta. TarOi-Nom selfDi/selfNi-Gen hometown-Dat returned 'TarOi returned to hiSj own hometown.'
d. TarOj-ga ??zisini/zibunrga baka-da to omotteiru. TarorNom self^/self^-Nom fool-be that think 'TarOi tliinks that he; himself is a fool.'
I would like to claim, however, that the oddness observed in (86) has nothing to do with the syntactic component of Japanese grammar, but it is rather related to the following two factors:
(87) a. Distinction between a formal style and an informal style b. Distinction between an honorific style and anon-honorific style
More precisely, my claim is that zisin is a free morpheme, but can only be used in a formal style, or in an honorific style. The first factor can be shown by the contrast between (88) and (89), where the former is in an informal style, and the latter, in a formal style:
(88) a. Anone, Tarorga ne *zisini/zibunro naguttan datte sa. say TarOj-Nom you.know selfDi/selfNi-Acc hit I.hear you.know 'Say, I hear TarOj hit himself^'
b. Anone, TarOj-ga ne *zisini/zibunro kenasitan datte sa. say TarOi-Nom you.know selfc/self^-Acc criticized I.hear you.know 'Say, I hear Taro; criticized himself^'
c. Anone, Tarorga ne *zisin/zibunrno kokyoo-ni kaettan datte sa. say TarOi-Nom you.know self^/self^-Gen hometown-Dat returned I.hear you.know 'Say, I hear TarOj went back to hiSj hometown.'
d. Anone, Taro,-ga ne *zisin,/zibunrga baka-datte omottcrun datte sa. say TarOj-Nom you.know self,)i/.sclfN,-Noin fool-is think I.hear you.know 'Say, I hear Taro; tliinks that he; is an idiot.'
(89) a. Tanakapga zisnVzibutyo ooda-suru koto-wa yurusareru koto dewa-nakatta. Tanaka-Nom selfDi/selfNi-Acc hit-do that-Top be.permitted tiling not.was 'It was prohibited that Tanakaj hit himself;.'
(56) H JEM
b. Tanakaj-ga zisin/zibuivo hihan-suru koto-ga hituyoo-to-sareteita. Tanakaj-Nom self^/self^-Acc criticize-do (hat-Nom it.was.necessary 'It was necessary that Tanakaj criticized himself-.'
c. Tanakaj-ga zisini/zibunrno kokyoo-ni kaette-simatta koto-o daremo-ga Tanakaj-Nom seIfDi/selfNi-Gen hometown-Dat went.back that-Acc everyone-Nom kooteetekini hyooka-siyoo-to-siteita.
positively was.trying.to.evaluate
"The fact that Tanakaj went back to hisf hometown, everyone was trying to evaluate itpositively.'
d. Tanakaj-ga zisuyzibutvga titeki-de-nai koto-o ninsiki-siteiru. Tanakaj-Nom selfDi/selfNj-Nom intellectual-be-not that-Acc recognition-do 'Tanaka^ recognizes that he, is unintellectual.'
Now let us turn to the second factor. Many Japanese nouns, adjectives, and adverbs can be turned into honorific forms or polite forms by prefixing o- or go-:
(90) a. [Noun] (91) a. [Noun]
kuti > o-kuti kainin -> go-kainin
'mouth' 'pregnancy'
b. [Adjective] b. [Adjective]
utukusyii > o-utukusyii rippana -> go-rippana
'beautiful' 'splendid'
c. [Adverb] c. [Adverb]
hayaku-> o-hayaku
yukkuri -> go-yukkuri
'fast' 'slowly'
The two kinds of anaphors in Japanese also can be turned into honorific forms: (92) a. zibun->go-zibun
b. zisin->go-zisin
I will assume that o/go-prefixation takes place in the lexicon, and that the only contribution of the prefix is to add the honorific meaning to the base, preserving all the syntactic and semantic information of zisin. This is indeed the case, as can be seen in (93):
(93) a. [cptNpTanaka-sensei] go-zisin] [dp [npTanaka-teacher] Hon-selfD
b. * tiisana go-zisin/ *kinoo-no go-zisin small Hon-selfD yesterday-Gen Hon-selfD c. *go-zisin-ra
Hon-selfD-Pl
The examples in (93) show the following:(i) go-zisin can take an NP-complement, (ii) go-zisin cannot be modified by adjectives or possessives, and (iii) go-zisin cannot be suffixed with a plural morpheme. Therefore, it is reasonable to assume that we can use go-zisin in order to investigate the properties of zisin. If this assumption is correct, we have a very good probe into the behavior of zisin. This is because, while the use of zisin is constrained by the stylistic factor which is not so easy to control sometimes, we can easily construct sentences in which go-zisin is well-formed. Compare the sentences in (94) with those in (86):
